Prediction - script API

StackState Self-hosted v4.6.x

This page describes StackState version 4.6.

Go to the documentation for the latest StackState release.

Function: predictMetrics

Predict metrics for any metric query coming from any data source.

Args:

• predictorName - the prediction preset to use:

  • fft - Fast Fourier Transformation. Used for seasonal metrics with a repeatable pattern that has a cycle.

  • linear - Used for metrics that have a linear trend.

  • hmn - Harmonic Mean Normal. Used in difficult cases when metrics are not clearly seasonal or linear.

• horizon - How much future to predict. The horizon is specified in duration format.

• query - The metrics to use for the prediction. A telemetry query that returns metrics, created using the Telemetry.query() function followed by .compileQuery().

Builder methods:

• predictionPoints(points: Int) - the number of points to the horizon.

• includeHistory(start?: Instant, end?: Instant) - call this builder method to include the result of the query in the return value. Optionally a start and end can be added to limit the included history using the instant format. When not specifying the start and end the whole history will be included.

Return type:

AsyncScriptResult[PredictionResponse]

A PredictionResponse, which contains the following fields:

Fields:

• PredictionResponse.request - the request made to the prediction API of type PredictionRequest.

• PredictionResponse.history - optional, the history used for prediction of type MetricTelemetry. Empty if .includeHistory() was not used.

• PredictionResponse.prediction - the predicted metrics.

The PredictionRequest type has the following fields:

• PredictionRequest.query - the query provided to predictMetrics.

• PredictionRequest.predictor - the name and configuration of the predictor.

• PredictionRequest.horizon - the prediction horizon.

• PredictionRequest.predictionPointCount - the number of predicted points.

• PredictionRequest.historyResponse - options of the history response.

The MetricTelemetry has the following fields:

• MetricTelemetry.result.data - the two dimensional array with values and time stamps.

Examples:

Predict 8 points of disk utilization for myHost for the next four hours based on the last four weeks of data:

Prediction.predictMetrics("linear", "4h",
    Telemetry.query("MyDatasource", 'host="myHost" and name="diskutil"')
        .metricField("value")
        .start("-4w")
        .compileQuery()
).includeHistory().predictionPoints(8)